Image Forgery Detection and Classification using Deep Learning
Davinder Paul Singh, Vinith Kumar Nair, Ram Singh, Puneet Bafna, M. Vedaraj, Veeraraghavan Vishnu Priya · 2024
This research employs an unsupervised learningbased autoencoder and decoder to identify picture fraud. Individuals utilize the internet to upload and share photographs on various social media platforms. This project ensures social security by detecting counterfeit photographs on internet-based social media platforms. Images undergo color illumination processing and are then transformed into positive and negative patches. The patches are saved in a.npy array with dimensions of $30 \times 30 \times 3$. The Autoencoder trains on patches classified as true positives and negatives. An auto decoder utilizes the minimum necessary pixels to reassemble pictures. The auto-encoder employs a non-linear modification to decrease the dimensionality. The color-illuminated photos were utilized in Harrie’s corner detection machine learning (ML) algorithm. Initially, it converts the input into simple signals. Next, it consists of several convolution layers, followed by an added that combines output three and output four, along with max pooling. Next, it reduces the resolution of the supplied image to its highest level of compression. An auto decoder recreates pictures by utilizing the minimal and necessary pixels. It can duplicate the output picture onto the input image but with reduced quality. The architecture consists of many convolutional layers, combined using an added that takes the outputs from the third and fourth layers and performs up sampling. Experiments were conducted on many publicly accessible databases, including BSDS300, CASIA v1.0, CASIA v2.0, and COMOFOD.